運用專利分析探討自然語言處理技術之發展

Abstract

在當今知識經濟的背景底下，隨著時代背景的轉變以及深度學習的興起，促使神經網路架構在自然語言處理（Natural Language Processing）技術中獲得突破性的成果，也再度燃起人們對於實現人機通訊的興趣。目前，NLP技術已廣泛應用於文句的情緒分析、詞類標示、機器翻譯、文法錯誤更正及語音辨識等各種使用情境，因此可將該技術視為是語言學與AI領域中備受矚目的技術，為此追蹤NLP技術之投入量變化、相關研發單位的專利布局，已是研發人員最需熱切探究與關心的議題。專利為各國企業的無形資產，不僅可以進行新技術的保護外，它同時也是一份完整描述專利技術內涵的紀錄文件。專利分析除了可以用來窺視整體趨勢的動向，以獲得確切的情報來保障企業組織的研發計畫方向及成果的價值性外，也可以通過時刻關注競爭對手和全球的技術研發動向，達到具備經濟效益的專利布局。為此本研究通過分析USPTO專利資料庫，窺視1976至2020年期間NLP技術之專利數量變化以及相關單位的投入狀況。經分析後發現NLP技術的專利申請數量成長快速，且該領域最常被引用之文獻主題以智能助理、語音識別居多，所以相關研究者可優先閱覽這類文獻來瞭解NLP技術的現況。Partnership on AI的董事會成員皆為該領域主要的專利權人，其中又以IBM（19.05%）位居首位，相較之下，臺灣專利權人雖以工研院、資策會為主，但歷年專利件數皆少於3件，顯示相關單位並未在美國專利市場上有重大的技術布局。反觀中國雖然在申請美國專利的腳步上比其他境外國家晚，但自2011年起便急起直追，並於2018年超越美國主要之境外申請國日本，凸顯中國對於技術布局與智慧財產的重視。In the knowledge-based economy, with the change of the times and the rise of deep learning, the neural network architecture has made a breakthrough in natural language processing technology, which has rekindled people's interest in realizing man-machine communication. At present, NLP technology has been widely used in emotion analysis, part-of-speech tagging, machine translation, grammar error correction and speech recognition, so NLP technology can be regarded as an attractive technology in linguistics and artificial intelligence. Therefore, tracking the change of NLP technology investment and the patent layout of relevant R&D units has become the most urgent problem that R&D personnel explore and care about. Patents are intangible assets of enterprises in various countries, which can not only protect new technologies, but also provide a complete record document describing the connotation of patented technologies. Patent analysis can not only be used to peek into the overall trend, so as to obtain accurate information and ensure the direction of R&D plans and the value of each enterprises's achievements, but also pay close attention to the technological R&D trends of competitors and the world at any times, so as to realize the economically favorable patent layout.Therefore, by analyzing the USPTO patent database, this study looks at the changes in the number of patents of NLP technology and the investment status of relevant units from 1976 to 2020. After patent analysis, it is found that the number of patent applications for NLP technology is increasing rapidly, and the most frequently cited literature topics in this field are intelligent assistants and speech recognition. Therefore, relevant researchers can read this kind of literature first to understand the current situation of NLP technology. The members of the board of directors of Partnership on AI are the main patentees in this field, with IBM （19.05%） ranking first. In contrast, although the patentees in Taiwan Province Province are mainly Industrial Technology Research Institute and the Institute for Information Industry, the number of patents over the past years is less than 3, which indicates that the relevant units have not attached importance to the technical layout in the US patent market. On the other hand, although China lags behind other overseas countries in applying for US patents, it has been catching up since 2011 and surpassed Japan, the major overseas applicant of the United States in 2018, which highlights China's emphasis on technology layout and intellectual property rights.